search

PRBonn / lidar-bonnetal

Semantic and Instance Segmentation of LiDAR point clouds for autonomous driving
http://semantic-kitti.org
MIT License
959 stars 206 forks source link

Early stopping #65

Closed jespadinha closed 4 years ago

jespadinha commented 4 years ago

Is early stopping being performed in this network? Or do they just save the latest best model checkpoint but still resume trainning until the end?

tano297 commented 4 years ago

Early stopping is "implemented" in the sense that only the best validation model is saved, but real early stopping which stops the training completely is hard to implement. I had this in at some point, but realized that finding an exit criteria in a noisy loss/accuracy metric is not trivial, so only the best validation model is saved and the stopping per se is left to you when you see the model really converged

jespadinha commented 4 years ago

Thank you for the answer!